2 * Copyright (C) 2003 Sistina Software
3 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
5 * This file is released under the LGPL.
8 #include <linux/init.h>
9 #include <linux/slab.h>
10 #include <linux/module.h>
11 #include <linux/vmalloc.h>
12 #include <linux/dm-io.h>
13 #include <linux/dm-dirty-log.h>
15 #include <linux/device-mapper.h>
17 #define DM_MSG_PREFIX "dirty region log"
19 struct dm_dirty_log_internal {
20 struct dm_dirty_log_type *type;
22 struct list_head list;
26 static LIST_HEAD(_log_types);
27 static DEFINE_SPINLOCK(_lock);
29 static struct dm_dirty_log_internal *__find_dirty_log_type(const char *name)
31 struct dm_dirty_log_internal *log_type;
33 list_for_each_entry(log_type, &_log_types, list)
34 if (!strcmp(name, log_type->type->name))
40 static struct dm_dirty_log_internal *_get_dirty_log_type(const char *name)
42 struct dm_dirty_log_internal *log_type;
46 log_type = __find_dirty_log_type(name);
48 if (!log_type->use && !try_module_get(log_type->type->module))
63 * Attempt to retrieve the dm_dirty_log_type by name. If not already
64 * available, attempt to load the appropriate module.
66 * Log modules are named "dm-log-" followed by the 'type_name'.
67 * Modules may contain multiple types.
68 * This function will first try the module "dm-log-<type_name>",
69 * then truncate 'type_name' on the last '-' and try again.
71 * For example, if type_name was "clustered-disk", it would search
72 * 'dm-log-clustered-disk' then 'dm-log-clustered'.
74 * Returns: dirty_log_type* on success, NULL on failure
76 static struct dm_dirty_log_type *get_type(const char *type_name)
78 char *p, *type_name_dup;
79 struct dm_dirty_log_internal *log_type;
84 log_type = _get_dirty_log_type(type_name);
86 return log_type->type;
88 type_name_dup = kstrdup(type_name, GFP_KERNEL);
90 DMWARN("No memory left to attempt log module load for \"%s\"",
95 while (request_module("dm-log-%s", type_name_dup) ||
96 !(log_type = _get_dirty_log_type(type_name))) {
97 p = strrchr(type_name_dup, '-');
104 DMWARN("Module for logging type \"%s\" not found.", type_name);
106 kfree(type_name_dup);
108 return log_type ? log_type->type : NULL;
111 static void put_type(struct dm_dirty_log_type *type)
113 struct dm_dirty_log_internal *log_type;
119 log_type = __find_dirty_log_type(type->name);
123 if (!--log_type->use)
124 module_put(type->module);
126 BUG_ON(log_type->use < 0);
132 static struct dm_dirty_log_internal *_alloc_dirty_log_type(struct dm_dirty_log_type *type)
134 struct dm_dirty_log_internal *log_type = kzalloc(sizeof(*log_type),
138 log_type->type = type;
143 int dm_dirty_log_type_register(struct dm_dirty_log_type *type)
145 struct dm_dirty_log_internal *log_type = _alloc_dirty_log_type(type);
152 if (!__find_dirty_log_type(type->name))
153 list_add(&log_type->list, &_log_types);
162 EXPORT_SYMBOL(dm_dirty_log_type_register);
164 int dm_dirty_log_type_unregister(struct dm_dirty_log_type *type)
166 struct dm_dirty_log_internal *log_type;
170 log_type = __find_dirty_log_type(type->name);
181 list_del(&log_type->list);
188 EXPORT_SYMBOL(dm_dirty_log_type_unregister);
190 struct dm_dirty_log *dm_dirty_log_create(const char *type_name,
191 struct dm_target *ti,
192 unsigned int argc, char **argv)
194 struct dm_dirty_log_type *type;
195 struct dm_dirty_log *log;
197 log = kmalloc(sizeof(*log), GFP_KERNEL);
201 type = get_type(type_name);
208 if (type->ctr(log, ti, argc, argv)) {
216 EXPORT_SYMBOL(dm_dirty_log_create);
218 void dm_dirty_log_destroy(struct dm_dirty_log *log)
224 EXPORT_SYMBOL(dm_dirty_log_destroy);
226 /*-----------------------------------------------------------------
227 * Persistent and core logs share a lot of their implementation.
228 * FIXME: need a reload method to be called from a resume
229 *---------------------------------------------------------------*/
231 * Magic for persistent mirrors: "MiRr"
233 #define MIRROR_MAGIC 0x4D695272
236 * The on-disk version of the metadata.
238 #define MIRROR_DISK_VERSION 2
245 * Simple, incrementing version. no backward
253 struct dm_target *ti;
255 uint32_t region_size;
256 unsigned int region_count;
259 unsigned bitset_uint32_count;
260 uint32_t *clean_bits;
262 uint32_t *recovering_bits; /* FIXME: this seems excessive */
268 DEFAULTSYNC, /* Synchronize if necessary */
269 NOSYNC, /* Devices known to be already in sync */
270 FORCESYNC, /* Force a sync to happen */
273 struct dm_io_request io_req;
279 struct dm_dev *log_dev;
280 struct log_header header;
282 struct dm_io_region header_location;
283 struct log_header *disk_header;
287 * The touched member needs to be updated every time we access
288 * one of the bitsets.
290 static inline int log_test_bit(uint32_t *bs, unsigned bit)
292 return ext2_test_bit(bit, (unsigned long *) bs) ? 1 : 0;
295 static inline void log_set_bit(struct log_c *l,
296 uint32_t *bs, unsigned bit)
298 ext2_set_bit(bit, (unsigned long *) bs);
302 static inline void log_clear_bit(struct log_c *l,
303 uint32_t *bs, unsigned bit)
305 ext2_clear_bit(bit, (unsigned long *) bs);
309 /*----------------------------------------------------------------
311 *--------------------------------------------------------------*/
312 static void header_to_disk(struct log_header *core, struct log_header *disk)
314 disk->magic = cpu_to_le32(core->magic);
315 disk->version = cpu_to_le32(core->version);
316 disk->nr_regions = cpu_to_le64(core->nr_regions);
319 static void header_from_disk(struct log_header *core, struct log_header *disk)
321 core->magic = le32_to_cpu(disk->magic);
322 core->version = le32_to_cpu(disk->version);
323 core->nr_regions = le64_to_cpu(disk->nr_regions);
326 static int rw_header(struct log_c *lc, int rw)
328 lc->io_req.bi_rw = rw;
330 return dm_io(&lc->io_req, 1, &lc->header_location, NULL);
333 static int read_header(struct log_c *log)
337 r = rw_header(log, READ);
341 header_from_disk(&log->header, log->disk_header);
343 /* New log required? */
344 if (log->sync != DEFAULTSYNC || log->header.magic != MIRROR_MAGIC) {
345 log->header.magic = MIRROR_MAGIC;
346 log->header.version = MIRROR_DISK_VERSION;
347 log->header.nr_regions = 0;
350 #ifdef __LITTLE_ENDIAN
351 if (log->header.version == 1)
352 log->header.version = 2;
355 if (log->header.version != MIRROR_DISK_VERSION) {
356 DMWARN("incompatible disk log version");
363 static int _check_region_size(struct dm_target *ti, uint32_t region_size)
365 if (region_size < 2 || region_size > ti->len)
368 if (!is_power_of_2(region_size))
374 /*----------------------------------------------------------------
375 * core log constructor/destructor
377 * argv contains region_size followed optionally by [no]sync
378 *--------------------------------------------------------------*/
380 static int create_log_context(struct dm_dirty_log *log, struct dm_target *ti,
381 unsigned int argc, char **argv,
384 enum sync sync = DEFAULTSYNC;
387 uint32_t region_size;
388 unsigned int region_count;
389 size_t bitset_size, buf_size;
392 if (argc < 1 || argc > 2) {
393 DMWARN("wrong number of arguments to dirty region log");
398 if (!strcmp(argv[1], "sync"))
400 else if (!strcmp(argv[1], "nosync"))
403 DMWARN("unrecognised sync argument to "
404 "dirty region log: %s", argv[1]);
409 if (sscanf(argv[0], "%u", ®ion_size) != 1 ||
410 !_check_region_size(ti, region_size)) {
411 DMWARN("invalid region size %s", argv[0]);
415 region_count = dm_sector_div_up(ti->len, region_size);
417 lc = kmalloc(sizeof(*lc), GFP_KERNEL);
419 DMWARN("couldn't allocate core log");
425 lc->region_size = region_size;
426 lc->region_count = region_count;
430 * Work out how many "unsigned long"s we need to hold the bitset.
432 bitset_size = dm_round_up(region_count,
433 sizeof(*lc->clean_bits) << BYTE_SHIFT);
434 bitset_size >>= BYTE_SHIFT;
436 lc->bitset_uint32_count = bitset_size / sizeof(*lc->clean_bits);
442 lc->clean_bits = vmalloc(bitset_size);
443 if (!lc->clean_bits) {
444 DMWARN("couldn't allocate clean bitset");
448 lc->disk_header = NULL;
451 lc->log_dev_failed = 0;
452 lc->header_location.bdev = lc->log_dev->bdev;
453 lc->header_location.sector = 0;
456 * Buffer holds both header and bitset.
458 buf_size = dm_round_up((LOG_OFFSET << SECTOR_SHIFT) +
459 bitset_size, ti->limits.hardsect_size);
461 if (buf_size > dev->bdev->bd_inode->i_size) {
462 DMWARN("log device %s too small: need %llu bytes",
463 dev->name, (unsigned long long)buf_size);
468 lc->header_location.count = buf_size >> SECTOR_SHIFT;
470 lc->io_req.mem.type = DM_IO_VMA;
471 lc->io_req.notify.fn = NULL;
472 lc->io_req.client = dm_io_client_create(dm_div_up(buf_size,
474 if (IS_ERR(lc->io_req.client)) {
475 r = PTR_ERR(lc->io_req.client);
476 DMWARN("couldn't allocate disk io client");
481 lc->disk_header = vmalloc(buf_size);
482 if (!lc->disk_header) {
483 DMWARN("couldn't allocate disk log buffer");
484 dm_io_client_destroy(lc->io_req.client);
489 lc->io_req.mem.ptr.vma = lc->disk_header;
490 lc->clean_bits = (void *)lc->disk_header +
491 (LOG_OFFSET << SECTOR_SHIFT);
494 memset(lc->clean_bits, -1, bitset_size);
496 lc->sync_bits = vmalloc(bitset_size);
497 if (!lc->sync_bits) {
498 DMWARN("couldn't allocate sync bitset");
500 vfree(lc->clean_bits);
502 dm_io_client_destroy(lc->io_req.client);
503 vfree(lc->disk_header);
507 memset(lc->sync_bits, (sync == NOSYNC) ? -1 : 0, bitset_size);
508 lc->sync_count = (sync == NOSYNC) ? region_count : 0;
510 lc->recovering_bits = vmalloc(bitset_size);
511 if (!lc->recovering_bits) {
512 DMWARN("couldn't allocate sync bitset");
513 vfree(lc->sync_bits);
515 vfree(lc->clean_bits);
517 dm_io_client_destroy(lc->io_req.client);
518 vfree(lc->disk_header);
522 memset(lc->recovering_bits, 0, bitset_size);
529 static int core_ctr(struct dm_dirty_log *log, struct dm_target *ti,
530 unsigned int argc, char **argv)
532 return create_log_context(log, ti, argc, argv, NULL);
535 static void destroy_log_context(struct log_c *lc)
537 vfree(lc->sync_bits);
538 vfree(lc->recovering_bits);
542 static void core_dtr(struct dm_dirty_log *log)
544 struct log_c *lc = (struct log_c *) log->context;
546 vfree(lc->clean_bits);
547 destroy_log_context(lc);
550 /*----------------------------------------------------------------
551 * disk log constructor/destructor
553 * argv contains log_device region_size followed optionally by [no]sync
554 *--------------------------------------------------------------*/
555 static int disk_ctr(struct dm_dirty_log *log, struct dm_target *ti,
556 unsigned int argc, char **argv)
561 if (argc < 2 || argc > 3) {
562 DMWARN("wrong number of arguments to disk dirty region log");
566 r = dm_get_device(ti, argv[0], 0, 0 /* FIXME */,
567 FMODE_READ | FMODE_WRITE, &dev);
571 r = create_log_context(log, ti, argc - 1, argv + 1, dev);
573 dm_put_device(ti, dev);
580 static void disk_dtr(struct dm_dirty_log *log)
582 struct log_c *lc = (struct log_c *) log->context;
584 dm_put_device(lc->ti, lc->log_dev);
585 vfree(lc->disk_header);
586 dm_io_client_destroy(lc->io_req.client);
587 destroy_log_context(lc);
590 static int count_bits32(uint32_t *addr, unsigned size)
594 for (i = 0; i < size; i++) {
595 count += hweight32(*(addr+i));
600 static void fail_log_device(struct log_c *lc)
602 if (lc->log_dev_failed)
605 lc->log_dev_failed = 1;
606 dm_table_event(lc->ti->table);
609 static int disk_resume(struct dm_dirty_log *log)
613 struct log_c *lc = (struct log_c *) log->context;
614 size_t size = lc->bitset_uint32_count * sizeof(uint32_t);
616 /* read the disk header */
619 DMWARN("%s: Failed to read header on dirty region log device",
623 * If the log device cannot be read, we must assume
624 * all regions are out-of-sync. If we simply return
625 * here, the state will be uninitialized and could
626 * lead us to return 'in-sync' status for regions
627 * that are actually 'out-of-sync'.
629 lc->header.nr_regions = 0;
632 /* set or clear any new bits -- device has grown */
633 if (lc->sync == NOSYNC)
634 for (i = lc->header.nr_regions; i < lc->region_count; i++)
635 /* FIXME: amazingly inefficient */
636 log_set_bit(lc, lc->clean_bits, i);
638 for (i = lc->header.nr_regions; i < lc->region_count; i++)
639 /* FIXME: amazingly inefficient */
640 log_clear_bit(lc, lc->clean_bits, i);
642 /* clear any old bits -- device has shrunk */
643 for (i = lc->region_count; i % (sizeof(*lc->clean_bits) << BYTE_SHIFT); i++)
644 log_clear_bit(lc, lc->clean_bits, i);
646 /* copy clean across to sync */
647 memcpy(lc->sync_bits, lc->clean_bits, size);
648 lc->sync_count = count_bits32(lc->clean_bits, lc->bitset_uint32_count);
651 /* set the correct number of regions in the header */
652 lc->header.nr_regions = lc->region_count;
654 header_to_disk(&lc->header, lc->disk_header);
656 /* write the new header */
657 r = rw_header(lc, WRITE);
659 DMWARN("%s: Failed to write header on dirty region log device",
667 static uint32_t core_get_region_size(struct dm_dirty_log *log)
669 struct log_c *lc = (struct log_c *) log->context;
670 return lc->region_size;
673 static int core_resume(struct dm_dirty_log *log)
675 struct log_c *lc = (struct log_c *) log->context;
680 static int core_is_clean(struct dm_dirty_log *log, region_t region)
682 struct log_c *lc = (struct log_c *) log->context;
683 return log_test_bit(lc->clean_bits, region);
686 static int core_in_sync(struct dm_dirty_log *log, region_t region, int block)
688 struct log_c *lc = (struct log_c *) log->context;
689 return log_test_bit(lc->sync_bits, region);
692 static int core_flush(struct dm_dirty_log *log)
698 static int disk_flush(struct dm_dirty_log *log)
701 struct log_c *lc = (struct log_c *) log->context;
703 /* only write if the log has changed */
707 r = rw_header(lc, WRITE);
716 static void core_mark_region(struct dm_dirty_log *log, region_t region)
718 struct log_c *lc = (struct log_c *) log->context;
719 log_clear_bit(lc, lc->clean_bits, region);
722 static void core_clear_region(struct dm_dirty_log *log, region_t region)
724 struct log_c *lc = (struct log_c *) log->context;
725 log_set_bit(lc, lc->clean_bits, region);
728 static int core_get_resync_work(struct dm_dirty_log *log, region_t *region)
730 struct log_c *lc = (struct log_c *) log->context;
732 if (lc->sync_search >= lc->region_count)
736 *region = ext2_find_next_zero_bit(
737 (unsigned long *) lc->sync_bits,
740 lc->sync_search = *region + 1;
742 if (*region >= lc->region_count)
745 } while (log_test_bit(lc->recovering_bits, *region));
747 log_set_bit(lc, lc->recovering_bits, *region);
751 static void core_set_region_sync(struct dm_dirty_log *log, region_t region,
754 struct log_c *lc = (struct log_c *) log->context;
756 log_clear_bit(lc, lc->recovering_bits, region);
758 log_set_bit(lc, lc->sync_bits, region);
760 } else if (log_test_bit(lc->sync_bits, region)) {
762 log_clear_bit(lc, lc->sync_bits, region);
766 static region_t core_get_sync_count(struct dm_dirty_log *log)
768 struct log_c *lc = (struct log_c *) log->context;
770 return lc->sync_count;
773 #define DMEMIT_SYNC \
774 if (lc->sync != DEFAULTSYNC) \
775 DMEMIT("%ssync ", lc->sync == NOSYNC ? "no" : "")
777 static int core_status(struct dm_dirty_log *log, status_type_t status,
778 char *result, unsigned int maxlen)
781 struct log_c *lc = log->context;
784 case STATUSTYPE_INFO:
785 DMEMIT("1 %s", log->type->name);
788 case STATUSTYPE_TABLE:
789 DMEMIT("%s %u %u ", log->type->name,
790 lc->sync == DEFAULTSYNC ? 1 : 2, lc->region_size);
797 static int disk_status(struct dm_dirty_log *log, status_type_t status,
798 char *result, unsigned int maxlen)
801 struct log_c *lc = log->context;
804 case STATUSTYPE_INFO:
805 DMEMIT("3 %s %s %c", log->type->name, lc->log_dev->name,
806 lc->log_dev_failed ? 'D' : 'A');
809 case STATUSTYPE_TABLE:
810 DMEMIT("%s %u %s %u ", log->type->name,
811 lc->sync == DEFAULTSYNC ? 2 : 3, lc->log_dev->name,
819 static struct dm_dirty_log_type _core_type = {
821 .module = THIS_MODULE,
824 .resume = core_resume,
825 .get_region_size = core_get_region_size,
826 .is_clean = core_is_clean,
827 .in_sync = core_in_sync,
829 .mark_region = core_mark_region,
830 .clear_region = core_clear_region,
831 .get_resync_work = core_get_resync_work,
832 .set_region_sync = core_set_region_sync,
833 .get_sync_count = core_get_sync_count,
834 .status = core_status,
837 static struct dm_dirty_log_type _disk_type = {
839 .module = THIS_MODULE,
842 .postsuspend = disk_flush,
843 .resume = disk_resume,
844 .get_region_size = core_get_region_size,
845 .is_clean = core_is_clean,
846 .in_sync = core_in_sync,
848 .mark_region = core_mark_region,
849 .clear_region = core_clear_region,
850 .get_resync_work = core_get_resync_work,
851 .set_region_sync = core_set_region_sync,
852 .get_sync_count = core_get_sync_count,
853 .status = disk_status,
856 static int __init dm_dirty_log_init(void)
860 r = dm_dirty_log_type_register(&_core_type);
862 DMWARN("couldn't register core log");
864 r = dm_dirty_log_type_register(&_disk_type);
866 DMWARN("couldn't register disk type");
867 dm_dirty_log_type_unregister(&_core_type);
873 static void __exit dm_dirty_log_exit(void)
875 dm_dirty_log_type_unregister(&_disk_type);
876 dm_dirty_log_type_unregister(&_core_type);
879 module_init(dm_dirty_log_init);
880 module_exit(dm_dirty_log_exit);
882 MODULE_DESCRIPTION(DM_NAME " dirty region log");
883 MODULE_AUTHOR("Joe Thornber, Heinz Mauelshagen <dm-devel@redhat.com>");
884 MODULE_LICENSE("GPL");